Dynamic Reference Sensing Scheme for Deeply Scaled STT-MRAM

Spin transfer torque magnetic random access memory (STT-MRAM) has been considered as a potential candidate for the next-generation nonvolatile memory. However, as technology continuously scales down, the sensing margin (SM) of STT-MRAM is significantly degraded because of the increased process variations and reduced supply voltage. Meanwhile the critical switching current of magnetic tunnel junction (MTJ) also reduces with technology scaling. The sensing current, which should be limited to prevent read disturbance (RD) during read operations, further degrades the SM. Therefore, the readability becomes a new challenge for the deeply scaled STT-MRAM. To alleviate this problem, various sensing circuits and schemes have recently been proposed. However, it is rather difficult to achieve a good tradeoff among the sensing reliability, latency, power and hardware efficiency etc. This paper presents a dynamic reference cell (DRC) as well as a dynamic reference sensing (DRS) scheme to deal with this problem. Monte-Carlo statistical simulations have been performed to show the superiority of the proposed DRS scheme compared with conventional sensing schemes.